NREGS in Bihar: preliminary findings

Report
Presentation at the Indian Statistical Institute Conference, New Delhi, December 2011
The impact on poverty
in Bihar of the National
Rural Employment
Guarantee Scheme
Puja Dutta, Rinku Murgai,
Martin Ravallion and
Dominique van de Walle
World Bank
India’s National Rural Employment
Guarantee Scheme (NREGS)
•
•
NREGS is the most ambitious effort India has ever
made to directly address absolute poverty.
Introduced in May 2006, the scheme:
–
–
–
–
guarantees 100 days/ h’hold/year of unskilled work on public
works projects in rural areas
provides work on demand after h’holds obtain a job card;
pays a piece-rate such that a normal worker can earn the
state-specific minimum wage rate set for the scheme
entitles women to equal wages
• Bihar has the highest
rural poverty rate of any
Indian state; 50+% in
2009/10. Also high % of
rural landless.
• Yet it has a relatively low
participation rate in
NREGS. Why?
Employment on NREGS (% of rural households)
NREGS should be important for Bihar
.7
r=0.13
.6
.5
.4
.3
.2
.1
Bihar
.0
10
15
20
25
30
35
40
45
50
55
60
Headcount index of rural poverty 2009/10
Bihar: highest rural poverty rate (56%) based on 2009/10 NSS
and PC poverty lines but one of lowest participation rates (0.22)
significantly below the regression line (t=-3.16).
3
How close does NREGS in Bihar come to
achieving its potential impact on poverty?
• Does NREGS really guarantee employment to anyone
who wants it in rural Bihar? Rationing of employment
opportunities may be common.
• forgone employment and income is unlikely to be zero in
reality; poor people cannot afford to be entirely idle, but
will find something to help make ends meet. But how
much forgone income?
• Leakage? Anecdotes that wages not paid in full; “ghost
workers;” contractors use machines and pocket the
difference. But how common in practice?
• How much impact on poverty in rural Bihar?
4
Outline of presentation
•
•
•
•
•
•
Demand for work and rationing
Wages
forgone income
Impacts on poverty
Comparisons with other states
Why so much rationing in poor
states?
• Main messages
5
Demand for work and the rationing
of the available work
6
A closer look at NREGS in Bihar using our
surveys
7
Survey data for the study
2009 Baseline survey (Round 1; “R1”)
• 3,000 randomly sampled households in 150 villages of
rural Bihar surveyed in April-July 2009
• 5,200 adult individuals, one male and one female from
each household
(“BREGS: The Movie:” A randomized control trial to test if
information can improve performance=> another paper)
2010 Follow-up survey (Round 2; “R2”)
• same villages/households in April-July 2010
8
Three groups
1.
2.
3.
Participants in NREGS
“Excess demanders”: those who say they would like to
work on NREGS but did not obtain work
The rest
% of Rural Households who:
participated in NREG
excess demanders
rest
% HH who desired participation
(participated or wanted to)
% of desired participants who:
got work in NREG
rationed
Round 1
Round 2
22.4
43.1
34.5
17.6
46.9
35.6
65.5
64.4
34.2
65.8
27.2
72.8
9
Does NREGS guarantee
employment? No
• Huge excess demand by men and women; 66% of
those households who wanted work on NREGS did not
get it in R1; rising to 73% in R2.
• Women are more likely to be excess demanders than
to be participants; 81% of women versus 62% of men
could not get NREGS work when they desired it in R1;
79% and 70% in R2.
10
Clear signs of pro-poor targeting
.2
.1
0
-.1
HH participation
.3
.4
Participation as a function of wealth
• Over the bulk of the
data (85%) participation
declines as wealth rises
• Poorest 15% are an
exception; amongst this
group, higher wealth
-5
0
5
10
implies higher
assets index
participation
with other HH controls and village characteristics
with other HH controls with village fixed effects
• But not just “selfwithout other HH controls
targeting” (whereby
non-poor don’t want the
Note: Wealth measured by an “assets index” as a
work)
function of housing and other assets and durables
11
.6
.7
.8
.9
1
1.1
Rationing as a function of wealth
-5
0
5
10
assets index
with other HH controls and village characteristics
with other HH controls with village fixed effects
without other HH controls
0
.2
.4
.6
.8
1
Rationing as a function of wealth
HH Rationing
• The probability of being
rationed (wanting NREGS
work but not getting it) is
lower for the poor.
• It is the relatively well-off
groups who are rationed
most in access to desired
work on NREGS.
HH Rationing
The rationing process is
also pro-poor
4
5
6
7
log(consumption per capita)
8
9
with other HH controls and village characteristics
with other HH controls with village fixed effects
without other HH controls
12
Findings from modeling participation and
demand for work on NREGS 1
• Demographics and gender matter
– Less likely to participate and more likely to be rationed in
family with
• more adult females
• female head
• Education matters
– Less well educated are more likely to want work on
NREGS and to get it
• Caste and religion matter
– Mahadalits more likely to both want NREGS work and to
get it; other SC more likely to be rationed out
– Hindus more likely to want work and to participate
13
Findings from modeling participation and
demand for work on NREGS 2
• Existing registration as “poor” and political connections
matter
– BPL card is a significant predictor of both demand for work and
participation
– Connections to Mukhiya, Sarpanch or block officer influence
whether work is obtained, but not demand for NREGS
• Village characteristics matter
– Mukhiya’s power influences participation in the scheme
– Better infrastructure (pucca road) decreases desired
participation but increases actual participation
– Higher inequality within village decrease demand for work on
NREGS and participation
14
In summary:
Substantial and systematic rationing
• Roughly two-thirds of those that want work don’t get it.
• Relatively well-off groups are more likely to be excluded
from access when they want it.
• There is a clear gender dimension to the rationing, in
that it is more likely for households with a large share of
adult women and female headed.
• Predominantly Muslim villages are more likely to be
rationed,
• Rationing more likely for those who do not have BPL
card and lack political connections.
• Those who are poor but lack the typical profile of the
poor appear to be more likely to be excluded from
15
access to the scheme when they want it.
Wages
16
NREGS is paying more than other
casual work
• In R1, PW wage 13% higher
than other casual work; 9% R2
• Wages of excess demanders
similar to the non-PW wages
of NREGS participants
• Those who get the jobs
on PW are essentially drawn
from the same wage
distribution as those who
do not get that work, but
want it.
• This is again suggestive
of rationing in the
assignment of jobs.
Density of daily wages
Casual work, Round 1
.04
.03
.02
.01
0
0
50
100
Rupees per day
150
200
wage rate public works
wage rate non-public works participant
wage rate non-public works excess demander
wage rate non-public works rest
17
Marked seasonality
In NREGS employment.
2
3
Rupees per day
120
100
80
60
40
1
Half the workers
earned less than
90% of the stipulated
wage rate
140
20
0
days(thousands)
Mean of actual wage
received on NREGS
is about 10% lower
than the stipulated
wage rate
10
Yet persistent gap between stipulated
wage and reported wage
01jul2008
01jan2009
01jul2009
01jan2010
01jul2010
survey
number of NREG days reported
stat. min. wage rate
reported NREG wage rate (female)
reported NREG wage rate (male)
18
Ratio of wage rate to minimum wage rate
Distribution in Round 1
A wage floor?
1
.8
.6
• In R1, only 14% of PW
workers earned less than
75% of the minimum wage
rate, as compared to 46%
of non-PW workers. (21%
and 45% in R2).
• It appears that NREGS is
able to provide a higher
(though still variable) “wage
floor” to participants than in
other casual work.
.4
.2
0
.5
1
ratio
1.5
NPW
PW
Ratio of wage rate to minimum wage rate
Distribution in Round 2
1
.8
.6
.4
.2
0
.5
1
1.5
ratio
NPW
PW
19
Less of a wage floor for women
• The proportion of women
earning considerably less
than the minimum wage
rate is markedly higher
than for men in both
rounds.
• In R1, 25% of women were
earning less than 75% of
the minimum wage rate, as
compared to 16% of men.
The gap only narrowed
slightly in R2.
Ratio of wage rate to minimum wage rate
Round 1
1
.8
.6
.4
.2
0
.5
1
Ratio
1.5
Male
Female
It is clear that NREGS is even less effective in providing
a wage floor for women than men.
20
But women get better wages under
NREG than casual market
Round 1
All
Men
Women
Round 2
All
Men
Women
Mean
St.dev.
Median
N
Public works
Other casual
labor
Public works
Other casual
labor
Public works
Other casual
labor
82.73
73.06
27.42
32.21
89.00
70.00
54
1050
85.81
77.87
25.23
30.51
89.00
77.14
41
829
73.03
45.01
32.62
24.24
80.00
42.00
13
221
Public works
Other casual
labor
Public works
Other casual
labor
Public works
Other casual
labor
94.03
86.16
23.89
42.68
100.00
97.50
119
806
98.11
97.44
21.34
39.83
100.00
100.00
76
577
86.81
57.74
26.58
35.90
100.00
50.00
43
229
Note: Missing identifiers entail that gender is unknown for some cases; hence, N is greater for “all” than
for the sum of men and women.
21
Leakage: Some comparisons with
administrative data
2008-09
Employment Outcomes
(million persondays)
Persondays generated
New employment created (net
of foregone employment)
Wage earnings (million Rs)
Total wages paid/owed
Total wages paid/received
Incremental wage gain (net of
wages foregone)
2009-10
Admin
data
Survey
Survey
as % of
admin
99
79
80%
114
98
86%
---
45
45%
---
57
50%
8396
8396
6310
5310
75%
63%
11087
11087
8910
7850
80%
71%
---
3371
40%
---
4715
43%
Admin
data
Survey
as % of
Survey admin
Memo:
% persondays foregone
43%
42%
% wages foregone
37%
40%
Note: net employment gain based on average foregone employment days of 43% and 42% in
R1 and R2. Net wage gain based on average foregone income of 37% and 40% in R1 and R2.
22
Forgone income
23
Participation often comes at a cost
• There is bound to be some loss of income from other
sources for at least some of those who take up public
works employment.
• Given that the wage rate is so much higher than that for
other work, some will naturally be attracted to NREGS
for the wage gain over alternative work.
• Others will no doubt be unable to find other work, and for
them the wage gain is also the net income gain from
NREGS.
• The literature on the impacts on poverty of public works
schemes has emphasized the importance of assessing
the forgone income.
24
Estimating forgone employment and
income
• Various approaches in the literature (time allocation
model; matching estimators).
• Here we use individual self-assessments.
– Individual knows a lot more than we do!
– However, counterfactual questions are not easy to answer
– Alternative interpretations of the question (=>)
• The survey asked counterfactual questions of actual
participants, to obtain their assessment of how many
days they expect to have worked and what they would
have earned in the absence on the program.
25
Alternative interpretations of selfassessed forgone income
• Interpretation 1: “Absence of the program for me only”
=> partial equilibrium forgone income
• Interpretation 2: “Absence of the program for everyone”
=> general equilibrium forgone income
• If respondents tended to use interpretation 1 then we will
probably overestimate forgone income to the extent that
counterfactual options overlap across respondents.
• Since this is a relatively new program we expect
interpretation 2 to apply to most people.
• Sensitivity tests to alternative interpretations.
26
Gross vs. net gain in employment
Round 1
• 79 million person days of employment.
• 43% mean forgone employment in R1.
Round 2
• 98 million person days of employment provided,
but 42% days had to be given up.
Gender: Slightly higher forgone employment for men than
women
27
Two alternatives to NREGS work
Lower mode at zero
1.5
2
Distribution of
the ratio of foregone income to PW wages
.5
1
Middle mode
at around 0.4
0
density
• In R1, 39% of those who
worked on PW had zero
forgone income (42% in R2).
• The rest reported that they
felt they would have been
working otherwise.
• Mean ratio of forgone income
to PW wages is 0.37 in R1,
rising to 0.39 in R2. The
corresponding medians are
0.36 and 0.31.
0
.2
.4
.6
.8
1
ratio
round1
round2
There is forgone income, but it varies considerably
between workers, which is not surprising.
28
Impacts on poverty
Bringing these elements together
29
Impact relative to “No-BREGS”
• The post-BREGS distribution of consumption is that
observed in the data
• The pre-BREGS distribution is derived from this by
subtracting the net gains from BREGS earnings, as given
by gross wages less the imputed forgone income.
• Ignoring assets and savings
30
More formally
• Actual (observed) post-BREGS poverty measure:
P(y1,….,yn; z)=P(y; z)
• Counterfactual poverty measure in the absence of
BREGS:
P(y – w + f; z)
where w = n-vector of actual wage earnings from BREGS
f = n-vector of forgone incomes due to taking up
BREGS work
• Impact on poverty =
P(y; z) – P(y – w + f; z)
31
Difference between cdf of consumption
before and after public works
About 1% point
reduction in poverty
due to the scheme
• A 1% point reduction in the
poverty rate at a poverty line
of slightly more than 5,000
Rupees per person per year.
• Amongst PW participants
alone, the impact is higher,
with a peak reduction in the
poverty rate of 3% points,
also at a poverty line of
5,000 Rupees.
Round 1
.01
0
-.01
-.02
-.03
0
5000
10000
15000
rupees per year
public workers
20000
25000
sample as a whole
Difference between cdf of consumption
before and after public works
Round 2
0
-.01
-.02
-.03
0
5000
10000
15000
rupees per year
public workers
20000
25000
32
sample as a whole
An idealized counterfactual
• What is the potential impact under ideal conditions?
–
–
–
–
100 days of work per household who wants that work
At the stipulated minimum wage rate for the scheme
Only fully unemployed people join, i.e., no forgone income
Again ignoring assets and savings
• Impact on poverty =
P(y – w + f; z) – P(y – w + f + 100w*d; z)
where d=(d1,…, dn) is desired participation (self-reported) in
NREG at stipulated minimum wage daily rate of w*
33
NREGS has a much larger potential
impact on poverty in Bihar
12% point reduction
in poverty rate
34
Rationing accounts for large share of
performance gap
• To estimate impact in the absence of rationing we:
– Scale up days of work to those desired, up to 100
days per h’hold with extra days valued at their mean
net wage.
– Give the median net earnings to those who wanted
work but did not get any.
• Then we get a poverty impact of 4-5% points.
35
An alternative
counterfactual
Difference between cdf of consumption
before and after BREGS (using administrative data)
Round 1
.005
0
-.005
• Suppose the same gross
expenditure on NREGS
was used to finance a
uniform lump-sum transfer
to all rural households,
whether poor or not.
• This would have a larger
impact on poverty than
NREGS as it is currently
performing.
• Also true of an alternative
using BPL card targeting.
-.01
-.015
-.02
0
5000
10000
15000
rupees per year
public workers
20000
25000
sample as a whole
Difference between cdf of consumption
before and after BREGS (using administrative data)
Round 2
.01
0
-.01
-.02
-.03
0
5000
10000
15000
rupees per year
public workers
20000
25000
36
sample as a whole
Comparisons with other states
Using 2009/10 NSS
37
• Participation rates on
NREGS across states
of India are only
weakly correlated
with poverty rates
across states.
• Why?
Employment on NREGS (% of rural households)
A puzzle about NREGS in India as a whole
.7
r=0.13
.6
.5
.4
.3
.2
.1
Bihar
.0
10
15
20
25
30
35
40
45
50
55
60
Headcount index of rural poverty 2009/10
38
• Poorer states have a
higher % of h’holds who
want work on NREGS
(actual employment +
those who say they want
work but could not get it).
• Though here too Bihar is
an outlier, with demand for
NREGS 0.14 below the
regression line (t=-2.58).
Demand for work on NREGS (% rural households 2009/10)
Yet poorer states of India have higher
demand for work on NREGS
.8
r=0.50
.7
.6
.5
Bihar
.4
.3
.2
.1
10
15
20
25
30
35
40
45
50
55
Headcount index of rural poverty 2009/10
39
60
Rationing in poorer states is the reason
.40
Share of rural households who were rationed
• Greater rationing—unmet demand for work on
the scheme—in some of
the poorest states.
• Highest rationing in
Bihar, but also high in
Jharkhand and Orissa.
• Low levels in TN, HP,
Rajasthan and Kerala
r=0.74
Bihar
.35
Jharkhand
Orissa
.30
Punjab
.25
Chhatisgarh
.20
.15
Kerala
.10
Rajasthan
Himachal Pradesh
Tamil Nadu
.05
10
15
20
25
30
35
40
45
50
55
60
Headcount index of rural poverty 2009/10
But low levels of rationing elsewhere, suggesting
that the scheme is working better
40
Why so much rationing in poor
states, including Bihar?
Here we can only offer some conjectures,
informed by the evidence and our field
observations
41
1. Lack of awareness on the part of
workers
• Our survey suggests that awareness of the right to work
is low, esp., women.
– 95% of men and 73% of women had heard about the program
– But most were unaware of their rights and entitlements under
NREGA. Low level of understanding about how to get work.
• Yes they still say they want work, but they don’t realize
they can demand work, and should get unemployment
benefit if it is not provided.
• But why are they so unaware? History of
subjugation/disempowerment, premised on illiteracy?
• Awareness is endogenous, but it can be influenced
externally: our RCT for the BREGS movie.
42
2. Low administrative capacity in poorest
states
• Supply side is slow to respond.
– Low levels of participation; few gram sabhas.
– Lags in execution; intermittant closures
– Poor flow of funds accounting
– Wages paid in cash not through POs
– Poor supervision
– Lack of transparency
• A scheme such as NREGS is likely to be harder to
implement in poor states.
43
3. Corruption?
• Surely corrupt local officials will have an incentive to
eliminate the rationing by starting more projects?
• Not if their own personal gain from doing so is
constrained by the design of the scheme.
• Corruption requires cooperation between a set of
stakeholders (officials and workers).
• Marginal cost of corruption may rise steeply at higher
levels of disbursement given checks and balances built
into the design.
• Very high MC when local officials would need to extend
their network of collusion beyond the “comfort zone” of
those they trust.
44
A simple model of rationing on NREGS
• Local officials maximize:
R(E)-C(E) s.t. E<D(w)
– R(E)=officials’ own revenue at
employment E; C(E) =cost of
corruption; D(w)=supply of labor
to NREGS at wage w
– Assume R”(E)<C”(E)
• There will be rationing in
equilibrium if E*<D(w), where
R’(E*)=C’(E*) .
MC
MB
E*
D(w)
D(w)-E*=Rationing
in equilibrium
45
Main messages
46
NREGS in Bihar is falling well short of its
potential impact on poverty
• Potential 12% point reduction in poverty in Bihar vs,
actual impact of 1% point.
• Some of this gap is hard to avoid, esp., forgone income
• But also many discrepancies between “theoretical ideal”
and practice
– Rationing is common; 2/3 of those who want work do not
get it
– While the rationing process is “pro-poor” overall some
socially/politically excluded groups have poor access
– Received wages lower than stipulated wages
– Worksites often lack facilities
– Processes weak
47
Performance issues are limiting the
potential benefits in Bihar
• Rationing makes it unlikely that there will be large
insurance and empowerment benefits.
• Shocks do not predict participation.
• Lack of awareness of rights under NREGS also makes it
unlikely that there would be large impacts on
empowerment.
• No sign of impacts on participation in village decisions or
that respect in the community improved.
48
Bihar is not typical of NREGS in India
• There is a collection of some of the poorest states
(Bihar, Orissa and Jharkhand) where rationing is
substantial.
• But also a number of states where this is not a problem,
suggesting that the scheme is likely to be working better
in reducing poverty and attaining its insurance and
empowerment potential.
• Harder to reduce poverty in poorer states.
49

similar documents